Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 08 1月, 2021 1 次提交
    • V
      net: bridge: notify switchdev of disappearance of old FDB entry upon migration · 90dc8fd3
      由 Vladimir Oltean 提交于 
      Currently the bridge emits atomic switchdev notifications for
dynamically learnt FDB entries. Monitoring these notifications works
wonders for switchdev drivers that want to keep their hardware FDB in
sync with the bridge's FDB.

For example station A wants to talk to station B in the diagram below,
and we are concerned with the behavior of the bridge on the DUT device:

                   DUT
 +-------------------------------------+
 |                 br0                 |
 | +------+ +------+ +------+ +------+ |
 | |      | |      | |      | |      | |
 | | swp0 | | swp1 | | swp2 | | eth0 | |
 +-------------------------------------+
      |        |                  |
  Station A    |                  |
               |                  |
         +--+------+--+    +--+------+--+
         |  |      |  |    |  |      |  |
         |  | swp0 |  |    |  | swp0 |  |
 Another |  +------+  |    |  +------+  | Another
  switch |     br0    |    |     br0    | switch
         |  +------+  |    |  +------+  |
         |  |      |  |    |  |      |  |
         |  | swp1 |  |    |  | swp1 |  |
         +--+------+--+    +--+------+--+
                                  |
                              Station B

Interfaces swp0, swp1, swp2 are handled by a switchdev driver that has
the following property: frames injected from its control interface bypass
the internal address analyzer logic, and therefore, this hardware does
not learn from the source address of packets transmitted by the network
stack through it. So, since bridging between eth0 (where Station B is
attached) and swp0 (where Station A is attached) is done in software,
the switchdev hardware will never learn the source address of Station B.
So the traffic towards that destination will be treated as unknown, i.e.
flooded.

This is where the bridge notifications come in handy. When br0 on the
DUT sees frames with Station B's MAC address on eth0, the switchdev
driver gets these notifications and can install a rule to send frames
towards Station B's address that are incoming from swp0, swp1, swp2,
only towards the control interface. This is all switchdev driver private
business, which the notification makes possible.

All is fine until someone unplugs Station B's cable and moves it to the
other switch:

                   DUT
 +-------------------------------------+
 |                 br0                 |
 | +------+ +------+ +------+ +------+ |
 | |      | |      | |      | |      | |
 | | swp0 | | swp1 | | swp2 | | eth0 | |
 +-------------------------------------+
      |        |                  |
  Station A    |                  |
               |                  |
         +--+------+--+    +--+------+--+
         |  |      |  |    |  |      |  |
         |  | swp0 |  |    |  | swp0 |  |
 Another |  +------+  |    |  +------+  | Another
  switch |     br0    |    |     br0    | switch
         |  +------+  |    |  +------+  |
         |  |      |  |    |  |      |  |
         |  | swp1 |  |    |  | swp1 |  |
         +--+------+--+    +--+------+--+
               |
           Station B

Luckily for the use cases we care about, Station B is noisy enough that
the DUT hears it (on swp1 this time). swp1 receives the frames and
delivers them to the bridge, who enters the unlikely path in br_fdb_update
of updating an existing entry. It moves the entry in the software bridge
to swp1 and emits an addition notification towards that.

As far as the switchdev driver is concerned, all that it needs to ensure
is that traffic between Station A and Station B is not forever broken.
If it does nothing, then the stale rule to send frames for Station B
towards the control interface remains in place. But Station B is no
longer reachable via the control interface, but via a port that can
offload the bridge port learning attribute. It's just that the port is
prevented from learning this address, since the rule overrides FDB
updates. So the rule needs to go. The question is via what mechanism.

It sure would be possible for this switchdev driver to keep track of all
addresses which are sent to the control interface, and then also listen
for bridge notifier events on its own ports, searching for the ones that
have a MAC address which was previously sent to the control interface.
But this is cumbersome and inefficient. Instead, with one small change,
the bridge could notify of the address deletion from the old port, in a
symmetrical manner with how it did for the insertion. Then the switchdev
driver would not be required to monitor learn/forget events for its own
ports. It could just delete the rule towards the control interface upon
bridge entry migration. This would make hardware address learning be
possible again. Then it would take a few more packets until the hardware
and software FDB would be in sync again.
Signed-off-by: NVladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: NNikolay Aleksandrov <nikolay@nvidia.com>
Reviewed-by: NIdo Schimmel <idosch@nvidia.com>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NJakub Kicinski <kuba@kernel.org>
      90dc8fd3
  3. 29 9月, 2020 1 次提交
    • N
      net: bridge: fdb: don't flush ext_learn entries · f2f3729f
      由 Nikolay Aleksandrov 提交于 
      When a user-space software manages fdb entries externally it should
set the ext_learn flag which marks the fdb entry as externally managed
and avoids expiring it (they're treated as static fdbs). Unfortunately
on events where fdb entries are flushed (STP down, netlink fdb flush
etc) these fdbs are also deleted automatically by the bridge. That in turn
causes trouble for the managing user-space software (e.g. in MLAG setups
we lose remote fdb entries on port flaps).
These entries are completely externally managed so we should avoid
automatically deleting them, the only exception are offloaded entries
(i.e. BR_FDB_ADDED_BY_EXT_LEARN + BR_FDB_OFFLOADED). They are flushed as
before.

Fixes: eb100e0e ("net: bridge: allow to add externally learned entries from user-space")
Signed-off-by: NNikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      f2f3729f
  5. 25 6月, 2020 3 次提交
  7. 05 11月, 2019 1 次提交
    • N
      net: bridge: fdb: eliminate extra port state tests from fast-path · 5d1fcaf3
      由 Nikolay Aleksandrov 提交于 
      When commit df1c0b84 ("[BRIDGE]: Packets leaking out of
disabled/blocked ports.") introduced the port state tests in
br_fdb_update() it was to avoid learning/refreshing from STP BPDUs, it was
also used to avoid learning/refreshing from user-space with NTF_USE. Those
two tests are done for every packet entering the bridge if it's learning,
but for the fast-path we already have them checked in br_handle_frame() and
is unnecessary to do it again. Thus push the checks to the unlikely cases
and drop them from br_fdb_update(), the new nbp_state_should_learn() helper
is used to determine if the port state allows br_fdb_update() to be called.
The two places which need to do it manually are:
 - user-space add call with NTF_USE set
 - link-local packet learning done in __br_handle_local_finish()
Signed-off-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      5d1fcaf3
  9. 02 11月, 2019 3 次提交
  11. 30 10月, 2019 7 次提交
  13. 31 5月, 2019 1 次提交
  15. 08 4月, 2019 1 次提交
    • N
      rhashtable: use bit_spin_locks to protect hash bucket. · 8f0db018
      由 NeilBrown 提交于 
      This patch changes rhashtables to use a bit_spin_lock on BIT(1) of the
bucket pointer to lock the hash chain for that bucket.

The benefits of a bit spin_lock are:
 - no need to allocate a separate array of locks.
 - no need to have a configuration option to guide the
   choice of the size of this array
 - locking cost is often a single test-and-set in a cache line
   that will have to be loaded anyway.  When inserting at, or removing
   from, the head of the chain, the unlock is free - writing the new
   address in the bucket head implicitly clears the lock bit.
   For __rhashtable_insert_fast() we ensure this always happens
   when adding a new key.
 - even when lockings costs 2 updates (lock and unlock), they are
   in a cacheline that needs to be read anyway.

The cost of using a bit spin_lock is a little bit of code complexity,
which I think is quite manageable.

Bit spin_locks are sometimes inappropriate because they are not fair -
if multiple CPUs repeatedly contend of the same lock, one CPU can
easily be starved.  This is not a credible situation with rhashtable.
Multiple CPUs may want to repeatedly add or remove objects, but they
will typically do so at different buckets, so they will attempt to
acquire different locks.

As we have more bit-locks than we previously had spinlocks (by at
least a factor of two) we can expect slightly less contention to
go with the slightly better cache behavior and reduced memory
consumption.

To enhance type checking, a new struct is introduced to represent the
  pointer plus lock-bit
that is stored in the bucket-table.  This is "struct rhash_lock_head"
and is empty.  A pointer to this needs to be cast to either an
unsigned lock, or a "struct rhash_head *" to be useful.
Variables of this type are most often called "bkt".

Previously "pprev" would sometimes point to a bucket, and sometimes a
->next pointer in an rhash_head.  As these are now different types,
pprev is NULL when it would have pointed to the bucket. In that case,
'blk' is used, together with correct locking protocol.
Signed-off-by: NNeilBrown <neilb@suse.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      8f0db018
  17. 19 1月, 2019 1 次提交
  19. 18 1月, 2019 1 次提交
    • P
      net: Add extack argument to ndo_fdb_add() · 87b0984e
      由 Petr Machata 提交于 
      Drivers may not be able to support certain FDB entries, and an error
code is insufficient to give clear hints as to the reasons of rejection.

In order to make it possible to communicate the rejection reason, extend
ndo_fdb_add() with an extack argument. Adapt the existing
implementations of ndo_fdb_add() to take the parameter (and ignore it).
Pass the extack parameter when invoking ndo_fdb_add() from rtnl_fdb_add().
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      87b0984e
  21. 17 12月, 2018 1 次提交
  23. 08 12月, 2018 1 次提交
  25. 18 10月, 2018 1 次提交
  27. 29 9月, 2018 1 次提交
  29. 13 9月, 2018 1 次提交
  31. 09 6月, 2018 1 次提交
  33. 04 5月, 2018 1 次提交
  35. 01 5月, 2018 1 次提交
  37. 01 2月, 2018 1 次提交
  39. 14 12月, 2017 1 次提交
    • N
      net: bridge: use rhashtable for fdbs · eb793583
      由 Nikolay Aleksandrov 提交于 
      Before this patch the bridge used a fixed 256 element hash table which
was fine for small use cases (in my tests it starts to degrade
above 1000 entries), but it wasn't enough for medium or large
scale deployments. Modern setups have thousands of participants in a
single bridge, even only enabling vlans and adding a few thousand vlan
entries will cause a few thousand fdbs to be automatically inserted per
participating port. So we need to scale the fdb table considerably to
cope with modern workloads, and this patch converts it to use a
rhashtable for its operations thus improving the bridge scalability.
Tests show the following results (10 runs each), at up to 1000 entries
rhashtable is ~3% slower, at 2000 rhashtable is 30% faster, at 3000 it
is 2 times faster and at 30000 it is 50 times faster.
Obviously this happens because of the properties of the two constructs
and is expected, rhashtable keeps pretty much a constant time even with
10000000 entries (tested), while the fixed hash table struggles
considerably even above 10000.
As a side effect this also reduces the net_bridge struct size from 3248
bytes to 1344 bytes. Also note that the key struct is 8 bytes.
Signed-off-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      eb793583
  41. 01 9月, 2017 1 次提交
  43. 30 8月, 2017 1 次提交
  45. 08 8月, 2017 1 次提交
  47. 04 7月, 2017 1 次提交
  49. 09 6月, 2017 3 次提交
  51. 01 5月, 2017 1 次提交
  53. 18 4月, 2017 1 次提交
  55. 25 3月, 2017 1 次提交